Journal of Machine Engineering, 2019, Vol. 19, No.

4, 70–81
ISSN 1895-7595 (Print) ISSN 2391-8071 (Online)

Received: 12 November 2019 / Accepted: 03 December 2019 / Published online: 20 December 2019

machine tool error, compensation,

control system

Pawel MEDYK1*
Jerzy JEDRZEJEWSKI1
Marcin KASPRZAK1

CNC MACHINE TOOL ERROR COMPENSATION SYSTEM IMPLEMENTATION

STRATEGIES AND THEIR CONSTRAINTS

This paper deals with the constraints imposed on error compensation systems by their implementation strategies.
Constraints stemming from the structure of the control system with an NCK architecture of the ADCBI type are
presented. Practical realizations of compensation and potential solutions offered by contemporary commercial
CNC controllers are discussed.

1. INTRODUCTION

Because of the strong competition on the CNC machine tool market, machine producers
need to continuously improve their machines. Therefore solutions improving selected
properties of machine tools, but without increasing their production costs, are sought.
A property which significantly improves machine tool competitiveness is its precision, i.e. its
ability to reproduce the user preset trajectory of tool motion with the possibly smallest error.
A machine tool error is usually understood as a difference between the actual tooltip center
point position and the point preset by the control program. This definition is a commonly used
simplification, very much justified because it directs one’s attention to the phenomena which
can cause machining precision to deteriorate, and to measures which should be taken to
improve this precision. The problem of a measure of machine tool precision is described in
more detail in [1].
Extensive research into methods of improving machine tool precision has been
conducted for years because of the general trend towards increasing the productivity
of machine tools and the demand for high-precision products. Generally, the methods
of improving machine tool precision can be divided into two groups: hardware methods and
software methods. The hardware methods include, i.a., design modifications, including
_____________
1
Wroclaw University of Science and Technology, Department of Machine Tools and Mechanical Technologies,
Wroclaw, Poland
*
E-mail: [email protected]
https://doi.org/ 10.5604/01.3001.0013.6231
P. Medyk et al./Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81 71

modifications of feed drive system [2], using higher quality assembly components and
equipping machine tools with additional position sensors for the realization of a hybrid loop
(Fig. 1). The latter uses motor shaft position measurements to achieve a fast system response,
and direct ball screw nut position measurements to ensure higher positioning precision [3].
Considering that the use of hardware solutions as a rule entails considerably higher machine
tool production costs and that improving precision by means of such methods is possible only
to a limited extent, software methods, such as error compensation systems and advanced feed
drive control algorithms [2], have been developed for many years to improve precision. The
fast technological progress, especially the growing computing power of computers (which
directly contributes to the development of advanced CNC control systems), is conducive to
the development of software methods. Thanks to the higher computing power of processors,
increasingly more advanced error compensation systems based on highly complex error
models can be implemented. This creates possibilities for the use of new error modelling
tools, e.g. convolutional neural networks (characterized by better knowledge generalization
than the commonly used methods [4]), and for creating systems performing optimization in
time closer to real time. Nevertheless, it is worth noticing that implementing error
compensation system on poorly designed machine with noticeable hysteresis may have
negative effects, such as deterioration of surface quality [5].

Fig. 1. Hybrid loop for controlling drive system in CNC machine tools [3]

By and large, contemporary control systems seem to be incapable of the compensation

of errors caused by changes in fast-variable quantities, such as the cutting force. Attempts
were made to compensate tool deformation and displacement caused by cutting forces by
predicting the force value in the given point using G-code and its later modification taking
into account the displacements caused by the cutting force [6]. A similar approach using
a G-code modification including corrections for errors caused by cutting forces was presented
in [7], where the cutting force was predicted on the basis of three feed and depth-of-cut
components. However, such methods do not realize the real-time compensation of errors
generated by the cutting force, but use error prediction regardless of the hardness of the stock
material, without taking into account random factors causing changes in the cutting force in
real time. Because of the partial randomness of the errors generated by the cutting force and
since flexible solutions, which could be used in small lot production and to improve many
different machines, are sought, research on compensation performed in real time or in time
possibly closest to real time is needed.
72 P. Medyk et al. /Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81

The present paper discusses the implementation of an error compensation system and
the associated difficulties stemming from the structure of the control system and the way it
operates. The NCK (Numerical Control Kernel) unit is briefly described and then the problem
of compensation implementation in the control system is discussed.

2. CONTROL SYSTEM STRUCTURE

Contemporary CNC control systems commonly use an NC system kernel of the ADCBI
type (acceleration/deceleration control before interpolation). A block diagram of such a kernel
is shown in Fig. 2. Thanks to the ADCBI motion planning architecture a deviation of the
actual tool path from the preset one can be avoided. In the case of ADCAI (acceleration/
deceleration control after interpolation) the deviation is due to the fact that the Acc/Dec
control is realized separately for each of the axes after the displacement which at the preset
feed rate is to occur in one interpolator clock pulse is determined. For NCK of the ADCBI
type the theoretical tool path generation error connected with changes in the feed rate amounts
to zero. But such an architecture is more difficult to implement than ADCAI and requires
greater computing power and more memory. Currently, NCK units have sufficient memory
and computing power reserves for ADCBI architecture implementation. The benefits
resulting from the use of this architecture include not only the elimination of the error
connected with changes in the velocity of the feed axes, but also the possibility of implemen-
ting additional functions, such as the Look Ahead algorithm.

Fig. 2. Block diagram of NC kernel of ADCBI type [3]

For better understanding of the difficulties in implementing error compensation systems

in contemporary CNC machine tools the particular ADCBI NCK modules and the way in
which data flow between them are briefly described below. This is also presented in Fig. 3.
This subject is dealt with in detail in [3].
P. Medyk et al./Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81 73

Fig. 3. Structure of NCK of ADCBI type with flow of data between modules [3]

After the NC program is executed it is interpreted by the control system. For this purpose
the interpreter module loads all the necessary system and user variables and then cyclically
reads the program block by block. For each of the blocks the interpreter performs appropriate
computations and transformations to determine the initial and target positions of the tooltip
central point in the basic coordinate system for the given block. Then it performs a tool radius
and length correction and checks if the calculated velocity and position values do not exceed
the allowable boundary values. Then the interpreter module writes the results in the buffer
used by the Look Ahead function (if the latter is implemented in the control system and is
turned on).
The Look Ahead algorithm significantly increases machining productivity when the
program consists of many short blocks (blocks describing small displacements). It is mainly
applicable to the machining of free-form surfaces since it is for the latter that the tool path is
usually described by a sequence of very short blocks. The aim of the Look Ahead algorithm
is to avoid a situation when the rate of feed is severely limited by the length of the blocks and
to prevent tool stoppage after the execution of each short block. The function executing the
algorithm draws the necessary information from many (from a few to a few hundred or even
a few thousand) blocks which follow the currently processed block and on this basis
determines the possibly highest feed rate for the given block.
74 P. Medyk et al. /Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81

For the processed block the number of iterations to be performed by the rough
interpolator in order to move to the given block’s preset end position and a table of values of
the (linear or angular) velocity with which the tool should move for the particular interpolator
iterations are determined in the acceleration control module. Determining speeds for the
successive interpolation iterations the acceleration controller attempts to bring the tool
acceleration as close as possible to the nearest maximum value dictated by the drives and/or
machine constraints.
The rough interpolator determines the target position for each interpolation iteration,
using information about the speed and the remaining distance or fragment of the arc
(considering that the acceleration, steady speed and deceleration phases must be realized in
a time being the multiple of the interpolator clock pulse time) and taking into account the
compensational displacements. The rough interpolator’s output is a table containing
information about the position to which the central tooltip point is to move in the given
interpolation clock pulse. This output is written in the buffer from which it is later retrieved
by the mapping-to-axis module.
After retrieving the data from the buffer the mapping-to-axis module determines the
displacement value for each of the axes in the given interpolation clock pulse and then sends
the data to the fine interpolator. Generally, the mapping-to-axis module can be regarded as
a part of the rough interpolator.
In general, the fine interpolator module prepares data for the position controller. The
preparation consists in dividing the distance which the given axis is to cover during an
interpolation clock pulse into a table of displacements of this axis in the next position
controller clock pulses. This is so because the position controller cycle time is many times
shorter than the interpolator cycle time.

3. ERROR COMPENSATION SYSTEM CONSTRAINTS DUE TO ADOPTED

IMPLEMENTATION STRATEGY

One of the ways in which an error compensation implementation strategy can be divided
is the division into on-line and off-line compensation. On-line compensation is performed
during the execution of the NC program within the particular NCK modules, whereas off-line
compensation is done prior to machining. In general, off-line strategies consist in
reconstructing the control program. Since the usability of off-line compensation methods is
not limited by the control system, only limitations in the implementation of on-line
compensation methods are discussed below.

3.1. ERROR COMPENSATION IN INTERPRETER

Since the interpreter writes data on the initial and end position of the whole block,
without determining the intermediate positions, the use of compensation in the interpretation
process results in only the determination of a new initial or end position of the block, with no
change in the position of the intermediate points determined in the rough interpolation
P. Medyk et al./Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81 75

module. Moreover, data are interpreted much ahead of time, stored in the buffer and then
processed by the next modules. The number of buffer elements varies depending on the length
of the blocks and can reach from a few to a few hundred or even a few thousand. For this
reason the time which passes from the instant a compensation correction is entered into the
interpreter module to the instant when the motion is executed is relatively long and impossible
to determine. Therefore entering compensation values into the interpreter module makes
sense only in the case of static or quasi-static errors, i.e. such whose value does not change or
changes very little in time and is independent of the position within a single block. If error
values change significantly between the block’s initial point and end point it is possible to
prepare earlier the G-code in such a way that the block will consist of many short (linear
of circular) segments. This, however, can significantly reduce machining efficiency,
especially in the case of NCK without the Look-Ahead module. The shorter the blocks, the
lower the machining efficiency.

3.2. ERROR COMPENSATION IN ROUGH INTERPOLATOR

Error compensation can be performed at the rough interpolation stage in order to

increase the precision of the compensation of position-dependent errors and to reduce the lag
between entering a correction and its execution. Owing to this solution, during the generation
of a position for the next interpolation clock pulses the displacement values determined for
this position by the compensation system can be taken into account. The distances between
the positions determined for the successive interpolation iterations are very small, whereby
position-dependent errors can be compensated with the possibly maximum precision
(the same with which the interpolator determines the successive nodes). Moreover, the rough
interpolator module processes blocks in advance and stores the processed data in the buffer,
whereby no so severe constraints connected with acting in real time as the ones occurring
within the fine interpolator are imposed on it. Unfortunately, the consequence is that the error
compensation performed as part of rough interpolation is asynchronous compensation.
The time passing from the writing of an instruction in the form of a preset position to
the instant when the motion described in this way is executed is indeterminate. This is due to
the fact that the number of instructions being already in the buffer from which the mapping
module and then the fine interpolator read data depends on very many factors and is not
constant. Consequently, also the execution time of all the instructions which have not been
executed yet, but are present in the buffer is variable and furthermore, much longer than the
interpolator cycle time.
Because the lag between the instant when compensation is introduced into the path and
the instant of its execution is unknown, it is not possible to introduce the compensation
of time-dependent errors if the changes do not occur slow enough. One can assume that for
time-dependent errors it makes sense to introduce compensation during rough interpolation if
the time in which a significant change in the error value occurs is greater than the product
of the interpolator cycle time and the buffer size (the maximum number of buffer elements
containing information about the position for one sampling cycle) between the rough
interpolator and the mapping module.
76 P. Medyk et al. /Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81

It is worth noting here that reducing the buffer size cannot be the solution to this
problem. The buffer enables the fluid motion of the tool if computations for the next block
are not completed during the interpolator clock pulse. If the buffer is too small, there is a risk
that before the interpolation of the next block is performed all the instructions present in the
buffer will be executed and the motion of the tool will be halted because of the absence
of execution instructions. The higher the rate of feed, the faster the buffer is emptied (at the
same interpolator clock pulse) and so the larger the buffer should be.
Each machine tool has certain kinematic properties which determine the maximum
allowable axis velocity values and the maximum acceleration and jerk values. This imposes
certain constraints on drive motion generation, which are taken into account in the interpreter
and acceleration controller modules.
As a result of the introduction of error compensation into rough interpolation the
determined tool displacement for the given interpolation iteration can differ from the one
which follows from the speed determined by the acceleration controller and is not subject to
verification by the modules preceding the interpolation. Therefore when implementing error
compensation systems one should bear in mind that there is a certain limitation on the value
of compensation corrections and one should check whether after the corrections the generated
motion does not result in the exceedance of the allowable velocity and acceleration values.
It is worth noting that both the compensation performed during the rough interpolator
clock pulse, by adding compensation corrections when determining the preset points for the
next interpolation iterations, and the compensation realized through a modification of the data
within the rough interpolator buffer are implementations of asynchronous on-line
compensation. The two implementations are characterized by almost the same lag between
the introduction of a compensation correction and the execution of the motion. Therefore,
simplifying one can assume that the two methods of implementation are methods of compen-
sation in rough interpolation.
The greatest difficulty in realizing compensation in the rough interpolator is that it is
necessary to modify the operation of the rough interpolator or to have access to the
interpolator buffer, which is not possible in the case of commercial CNC controllers. Thus
compensation in rough interpolation is possible only in open CNC systems [8] or using ready-
made functions of commercial CNC controllers.

3.3. ERROR COMPENSATION IN FINE INTERPOLATOR

Owing to compensation performed in the fine interpolator it would be possible to

compensate position- and time-dependent errors since compensation corrections would be
executed in the same interpolator clock pulse. The fine interpolator has no information about
the preset position, but only the preset displacement (in basic units of length) for the preset
axis. Therefore the compensation of position-dependent errors within this module would
require access to measurements of the tool position in the controllable axes. Such a solution,
shown in Fig. 4, was presented in [8].
At the same time it seems that the strongest barrier to the implementation of error
compensation in the fine interpolator is that the latter must operate in real time. Besides
P. Medyk et al./Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81 77

reading measurements off sensors, such a system would have to process the error model
during one interpolator clock pulse to determine the compensation value and then read data
off the shared memory (off which data are read by the position controller) in order to overwrite
them after corrections are added. Because of the constraints due to the necessity of performing
all the operations in one interpolator pulse, no solution of error compensation in fine
interpolation has been presented so far.

Fig. 4. Thermal error compensation in fine interpolation [8]

4. ERROR COMPENSATION STRATEGIES – SURVEY OF SOLUTIONS

4.1. OFF-LINE COMPENSATION

The most popular direction of research on error compensation is still off-line

compensation consisting in modifying G-code. This method can be used if the error is
repeatable or if it can be predicted. The most serious drawback of this method is that
compensation is effected through a change of the position of the program end blocks. Because
of this, the path has to be divided into short segments and arcs in order to achieve good results.
This can adversely affect machining efficiency. At the same time it is not possible to introduce
compensation in intermediate points between the block’s initial and end position. Nojodeh et
al. carried out the compensation of the geometric errors of a 3-axis milling machine, using the
method of dividing the nominal path into short segments and arcs [9], which resulted in
a 50% and 12% machining error reduction in two tests. Zhue et al. [10] compensated the
geometric errors of a 5-axis machine tool by modifying G-code consisting of short blocks,
which resulted in a 40-50% improvement in machining precision. Ratchew et al. [11]
proposed a strategy for the multilevel compensation of errors caused by the cutting force
78 P. Medyk et al. /Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81

during milling, through error prediction and G-code modification, showing that it is possible
to reduce the total machining error by as much as 70% using a single-level compensation
algorithm. The use of full multilevel compensation can yield even better results.

4.2. ON-LINE COMPENSATION

4.2.1. COMPENSATION IN INTERPRETER MODULE

Considering that compensation in the interpreter module generally comes down to block
end position modification, for time-invariable errors the compensation works the same as off-
line compensation and alike requires the program consisting of short blocks. The difference
between the two strategies is that compensation in the interpreter module makes it possible to
take into account the variation of errors over time.
The above approach was used by Vinod et al. [12] to realize the on-line compensation
of the positioning error of a turning lathe by introducing a compensation correction as a zero
point shift or a tool off-set (possible selection of a command). The cycle time for determining
a new correction value was set to 1 second. This solution did not use any external devices,
but only a program processing the error model on the same processing unit. This was made
possible by the open architecture of the CNC controller. A positioning error reduction of about
75% was achieved.
The same method of introducing compensation was used in [13] for thermal error
compensation and the error was reduced from 50 μm to 3 μm.

4.2.2. COMPENSATION IN ROUGH INTERPOLATOR

Compensation performed within rough interpolation can be effected by means of, e.g.,
the functions available in commercial CNC controllers, such as: temperature compensation,
sag compensation and the volumetric compensation system.
The error model used by the temperature compensation function [16] is stored in PLC.
At specified intervals (of a few minutes) PLC processes the temperature profile and performs
the linear approximation of the error function for the measured temperature. The function
approximating the error function is as follows:

𝐾𝑥 (𝑇) = 𝐾0 (𝑇) + 𝐾𝑃 (𝑇)(𝑃𝑋 − 𝑃0 )

where:
𝐾𝑥 (𝑃𝑥 , 𝑇) − temperature compensation value of axis at position 𝑃𝑥 and temperature 𝑇,
𝑃𝑋 − actual position of axis,
𝑃0 − reference position of axis,
𝐾𝑃 (𝑇) − coefficient for the position-dependent temperature compensation,
𝐾0 (𝑇) − position-independent temperature compensation value of axis.
Then linear function coefficients 𝐾0 and 𝐾𝑝 , determined through approximation, are sent
to NCK. As the target positions for the successive interpolation iterations are determined in
the rough interpolation process, the compensation correction value is computed. In order to
P. Medyk et al./Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81 79

avoid problems connected with a too high value of the compensation correction, the latter is
distributed over a few successive interpolator clock pulses. In this way a new position preset
for the given interpolation iteration is obtained. It is worth noting that if the temperature
compensation function is used, the asynchronous on-line compensation of position-dependent
errors is possible, provided they are successfully approximated in segments, with the linear
function. Also position-independent errors in the form of off-set can be compensated by
setting coefficient 𝐾𝑝 (𝑇) to zero.
The volumetric compensation system (VCS) [16] requires for its operation that all 21
geometric error components of a cartesian machine tool are determined for the whole
workspace. The data are written, in the form of tables, to a file from which they are read in
by the CNC controller at the program start. It is possible to create many files containing data
on the volumetric error for different temperatures and/or a different weight of the tool or the
workpiece.
The latest versions of VCS enable interpolation between compensation files (volumetric
error value determination for intermediate temperature or weight values) and compensation
which takes the direction of tool motion into account. After appropriate data are read in, the
volumetric compensation system computes compensation values which are taken into account
in the rough interpolation process [14]. However, error model of VCS cannot be modified
when the program is running. Hence this method can be used only to compensate static errors.
Compensation based on VCS can increase precision by as much as 75–80%.
The sag compensation function [15] operates on the basis of the preset compensation
table which contains pairs of the preset datum axis position and the compensational axis
compensation value. The function enables the use of various tables, depending on the
direction of motion of the given axis, and the selection of the same axis as simultaneously the
datum axis and the compensational axis. Similarly as in the case of VCS, the compensation
tables cannot be modified during machining, which means that only static errors can be
modified by means of this function.
Other methods of introducing compensation corrections into the interpolation process
are interpolatory corrections [15], such as: the external shift of the zero point, the DRF
(Differential Resolver Function) offset, and the on-line tool correction, available in contem-
porary CNC systems.
A strategy of implementing the compensation of position-dependent geometric errors
and thermal errors, based on the use of the DRF offset, was implemented by Cui et al. [16].
An error model was processed on an external device which after computing the compensation
value sent it to the CNC controller, imitating the hand-wheel signal. The experimental results
showed a geometric error reduction in axis x, y, w, of respectively 65%, 70% and 46%.
Shen et al. [8] presented a method of compensating static/quasi-static errors through the
modification of the data in the rough interpolator buffer. The compensation values were
determined on the basis of a compensation table which included: the preset position in the
given axis, the post-compensation position for forward motion and the post-compensation
position for backward motion. An attempt was made to verify the proposed method through
simulation and the method was found to correctly map the values contained in the
compensation tables in the modified rough interpolator data. However, this strategy can be
implemented only in open CNC systems.
80 P. Medyk et al. /Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81

As part of the above study also the compensation of the position-independent error was
carried out, using the PLC-CNC interface for the asynchronous determination of the off-set
by means of an inbuilt temperature compensation function (the method was discussed above).
The error reduction amounted to over 70%.

5. CONCLUSION

High precision is one of the main criteria which a modern machine tool must satisfy.
Striving to maximally exploit the machining precision potential of machine tools wide
research into software methods of improving machine tool precision is being conducted.
Contemporary control systems are capable of handling various error compensation system
implementation strategies. In this paper, compensation system constraints resulting from the
adopted compensation implementation strategy and whose source is the structure of the
control system were discussed.
The most popular error compensation strategy is the off-line method consisting in
G-code reconstruction. However, using this method one cannot take into account the effects
of changes in temperature and force-induced dynamically variable deformations on
machining precision. Only the end positions of program blocks can be corrected in this way.
The effect of changeable machining conditions can be taken into account through error
compensation in the interpreter module, but with a considerable lag. For this reason such
a strategy can be used for temperature compensation, but not for the compensation
of deformations induced by forces. A compensation correction can be introduced only by
modifying the block’s end position, which limits the achievable compensation precision.
Different strategies for implementing compensation in the rough interpolation module
were presented in this paper. Thanks to compensation performed within this module the
occurring changes can be responded to with a shorter delay than in the case of compensation
taking place in the interpreter and corrections can be introduced in the intermediate points
determined in the interpolation process.
So far no method of performing compensation in real time has been developed. Such
compensation could be realized in the fine interpolator module, but the contemporary CNC
controllers make this impossible. Further research in this area is needed.

ACKNOWLEDGEMENT

The authors express his thanks to NCBiR for supporting the research on intelligent machine tools, Project number
POIR.04.01.04-00-0067/17.

REFERENCES

[1] JÓZWIK J., 2018, Experimental methods of error identification in CNC machine tool operation, Lublin University
of Technology.
[2] ALTINTAS Y., VERL A., BRECHER C., URIARTE L., & PRITSCHOW G., 2011, Machine tool feed drives, CIRP
Annals, 60/2, 779–796.
[3] SUH S.H., 2008, Theory and design of CNC systems, Springer Science & Business Media.
P. Medyk et al./Journal of Machine Engineering, 2019, Vol. 19, No. 4, 70–81 81

[4] ZHANG X., YANG L., LOU P., et al., 2019, Thermal Error Modelling for Heavy Duty CNC Machine Tool Based
on Convolution Neural Network, IEEE 3rd Information Technology, Networking, Electronic and Automation
Control Conference ITNEC, IEEE, 665–669.
[5] SCHWENKE H., KNAPP W., HAITJEMA H., et al., 2008, Geometric error measurement and compensation of
machines – an update, CIRP Annals, 57/2, 660–675.
[6] SOORI M., AREZOO B., HABIBI M., 2014, Virtual machining considering dimensional, geometrical and tool
deflection errors in three-axis CNC milling machines, Journal of Manufacturing Systems, 33/4, 498–507.
[7] RAKSIRI C., PARNICHKUN M., 2004, Geometric and force errors compensation in a 3-axis CNC milling
machine, International Journal of Machine Tools and Manufacture, 44/12–13, 1283–1291.
[8] SHEN H., FU J., HE Y., YAO X., 2012, On-line asynchronous compensation methods for static/quasi-static error
implemented on CNC machine tools, International Journal of Machine Tools and Manufacture, 60, 14–26.
[9] NOJEDEH M.V., HABIBI M., AREZOO B., 2011, Tool path accuracy enhancement through geometrical error
compensation, International Journal of Machine Tools and Manufacture, 51/6, 471–482.
[10] ZHU S., DING G., QIN S., et al., 2012, Integrated geometric error modelling, identification and compensation
of CNC machine tools, International journal of machine tools and manufacture, 52/1, 24–29.
[11] RATCHEV S., LIU S., HUANG W., BECKER A.A., 2006, An advanced FEA based force induced error
compensation strategy in milling, International Journal of Machine Tools and Manufacture, 46/5, 542–551.
[12] VINOD P., REDDY T.N. SAJIN S., et al., 2014, Real-time positioning error compensation for a turning machine
using neural network, Procedia Materials Science, 5, 2293–2300.
[13] REDDY T.N., SHANMUGARAJ V., PRAKASH V., et al., 2014, Real-time thermal error compensation module
for intelligent Ultra Precision Turning Machine (iUPTM), Procedia materials science, 6, 1981–1988.
[14] TUREK P., JEDRZEJEWSKI J., MODRZYCKI W., 2010, Methods of machine tool error compensation, Journal
of Machine Engineering, 10/4, 5–25.
[15] Siemens Company publications, i.a.: Extended Functions, SINUMERIK 840D sl/840DE sl. Version 4.91, Extended
Functions, SINUMERIK 840D sl/828D, Version 4.7 SP2.
[16] CUI G., LU Y., GAO D., YAO Y., 2012, A novel error compensation implementing strategy and realizing on
Siemens 840D CNC systems, The International Journal of Advanced Manufacturing Technology, 61/5–8, 595–608.